SageMakerConfig

class autogluon.cloud.SageMakerConfig(*, region: str | None = None, role_arn: str | None = None, vpc_config: ~typing.Dict[str, ~typing.List[str]] | None = None, output_kms_key: str | None = None, volume_kms_key: str | None = None, tags: ~typing.Dict[str, str] = <factory>)[source]

Reusable SageMaker settings for predictors and foundation models.

Pass this as backend= to a cloud predictor or foundation model. Each object creates its own backend and jobs; sharing this config does not share execution state. Resource sizes and other operation settings remain named arguments to fit(), predict() and deploy().

Parameters:

  • region (str | None) – AWS region. If omitted, use the region in ~/.autogluon/cloud.yaml, then the boto3 default region.

  • role_arn (str | None) – SageMaker execution role ARN. If omitted, use the saved role, then the role of the current AWS identity.

  • vpc_config (Dict[str, List[str]] | None) – Networking for training jobs and models, as {"subnets": [...], "security_group_ids": [...]}.

  • output_kms_key (str | None) – KMS key for training artifacts, batch transform outputs, and repacked or cached model artifacts in S3.

  • volume_kms_key (str | None) – KMS key for training, batch transform and realtime endpoint storage volumes. Leave unset for instance types with local NVMe storage.

  • tags (Dict[str, str]) – Tags added to every SageMaker resource created by this backend.

Methods

Attributes

name

output_kms_key

region

role_arn

volume_kms_key

vpc_config

tags